| 2022-12-02 12:36:11 |
Rodolfo Alonso |
description |
This bug is related to https://bugs.launchpad.net/neutron/+bug/1578989.
The aim of this RFE is to improve LP#1578989. This feature implemented all the code necessary to allow Neutron to report the available network resources. Neutron was responsible of retrieving the administrator configuration and send it to Placement. Each node would be represented as a resource provider; the ML2 agents would be child RP resources and the physical network would be child RPs of the ML2 agents.
RP host 1
--> RP OVS agent
--> physnet 1: egress BW, ingress BW
This feature had a 1:1 representation of the physical networks available in each node (compute node). Each physical network had associated a physical device (a network interface) and this device had an associated egress and ingress bandwidth. We can see that as saying that "if we have a 10Gb NIC connected to physnet1, we'll assign those 10Gb for ingress/egress".
Currently we don't model the overlay networks (tunnelled networks). This is because there is no way to model the tunnelled networks. This RFE proposes to use a unique resource provider identifier to model the overlay networks in a host. Both in ML2/OVS and ML2/OVN (ML2/SR-IOV doesn't support tunnelled networks), the overlay traffic is sent to the "local_ip", that is the host IP address set on the local VTEP.
This "tunnelled resource provider" inventory will be requested for any port min-bw QoS rule that belongs to a tunnelled network (VXLAN, Geneve). |
This bug is related to https://bugs.launchpad.net/neutron/+bug/1578989.
The aim of this RFE is to improve LP#1578989. This feature implemented all the code necessary to allow Neutron to report the available network resources. Neutron was responsible of retrieving the administrator configuration and send it to Placement. Each node would be represented as a resource provider; the ML2 agents would be child RP resources and the physical network would be child RPs of the ML2 agents.
RP host 1
--> RP OVS agent
--> physnet 1: egress BW, ingress BW
This feature had a 1:1 representation of the physical networks available in each node (compute node). Each physical network had associated a physical device (a network interface) and this device had an associated egress and ingress bandwidth. We can see that as saying that "if we have a 10Gb NIC connected to physnet1, we'll assign those 10Gb for ingress/egress".
Currently we don't model the overlay networks (tunnelled networks). This is because there is no way to model the tunnelled networks. This RFE proposes to use a unique resource provider identifier to model the overlay networks in a host. Both in ML2/OVS and ML2/OVN (ML2/SR-IOV doesn't support tunnelled networks), the overlay traffic is sent to the "local_ip", that is the host IP address set on the local VTEP.
This "tunnelled resource provider" inventory will be requested for any port min-bw QoS rule that belongs to a tunnelled network (VXLAN, Geneve).
Related bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=2097932 |
|
